Summary

Pichai is defending a company whose stock has risen 4.5x to a roughly $2 trillion market cap under his tenure, while quarterly revenue grew from $20 billion to nearly $100 billion. His defense of Alphabet’s roughly $200 billion search-ad run rate is that generative AI expands demand rather than merely cannibalizing links. AI Overviews reach more than 1.5 billion users across 150-plus countries, produce sustained query growth where triggered, and will be joined by AI Mode, where queries are already two to three times longer than search queries two years ago. His operating rule: “The dilemma only exists if you treat it as a dilemma.”
The early economics weaken the sharpest bear case: AI Overview ads have reached the revenue baseline of traditional results, while serving cost has fallen dramatically over 18 months. Pichai says latency—not cost per query—is now the harder constraint because search users expect near-instant responses. His upside case is that “commercial information is also information,” so better AI should eventually improve ad relevance.
The standalone Gemini app trails on Friedberg’s cited scoreboard, but Pichai argues Alphabet’s AI distribution is broader than an app comparison. Friedberg cited figures from recent court testimony, including March data, of 350 million monthly Gemini users against 600 million for ChatGPT and 500 million for Meta AI; Pichai pointed to stronger engagement after Gemini 2.5 Pro and called it “still early days.” He emphasized usage across Search, YouTube, Cloud, Android-related Gemini experiences, and the Gemini app. Diversification matters too: YouTube and Cloud exited last year at a combined $110 billion.
Alphabet is spending $75 billion in 2025 to turn its full-stack infrastructure into both a cost advantage and a Cloud-capacity asset. Most capex goes to servers and data centers, while half of compute spending supports Google Cloud; seventh-generation TPUs and an Ironwood single pod above 40 exaflops underpin the claim that Google sits on the “Pareto frontier of performance and cost.” NVIDIA remains complementary: Gemini runs on GPUs as well as TPUs, and Pichai called NVIDIA’s software stack “world class.”
Pichai sees no fundamental model plateau yet, though harder gains should increasingly separate elite research teams. He described progress as “artificial jagged intelligence,” moving from pre-training to post-training, inference compute, and agentic workflows. DeepSeek forced an adjustment in priors about China’s proximity to the frontier, although Google’s internal comparison found Flash similarly efficient or arguably better.
Electricity and execution—not model theory—look like the binding constraints on AI-led growth. Friedberg cited US power capacity rising from roughly 1 to 2 terawatts by 2040, versus China moving from 3 to 8; Pichai acknowledged Google Cloud is already supply-constrained this year. Solar-plus-batteries, nuclear, geothermal, grids, permitting, transmission, and electrician shortages therefore become part of the Alphabet thesis.
Quantum, robotics, and ambient computing are moving from distant research options toward stated breakthrough windows. Pichai puts a useful quantum computation in roughly three to five years, a “magical moment” for robotics two to three years away, and compelling AR glasses a couple of product cycles out. Google previously tried the robotics application layer “too early”; Gemini’s vision-language-action models now change the premise.
The execution reset combines founder involvement, smaller teams, in-person intensity, and a portfolio unified by foundational technology rather than capital allocation. Sergey Brin is working with Gemini engineers on code, loss curves, architecture, and post-training, while Pichai has recreated Labs-style work with roughly 10-person teams and refocused employees on mission. He says Google is retaining critical AI talent and attracting top PhDs; Googlers have started more than 2,000 companies. Alphabet is “not a holding company” in the conventional sense—but Pichai still cited a heavily debated Netflix acquisition as one alternate path in the “multiverse.”

Deep dive

1. Google is rebuilding search around AI while defending a huge profit engine

Friedberg frames the disruption case against an extraordinary run: under Pichai, Alphabet’s stock has risen 4.5x to a roughly $2 trillion market cap, while quarterly revenue grew from $20 billion to nearly $100 billion. Search advertising runs near $200 billion against roughly $360 billion of total revenue, so moving too slowly risks losing users while moving too quickly could put the profit engine’s revenue at risk. The new chat paradigm supplies complete answers where classical search supplied links.
Pichai’s response begins nearly a decade earlier: Google Brain was underway in 2012, DeepMind was acquired in 2014, and he declared Google “AI first” after becoming CEO in 2015 because “AI is what will drive the biggest progress in search.” Transformers, BERT, and MUM subsequently improved core search quality.
The empirical case is AI Overviews, launched about a year earlier: more than 1.5 billion users in over 150 countries, broader query types, and continuing query growth wherever the feature triggers. AI Mode adds follow-up conversation and models that use Search as “a real, native tool”; its average query is already two to three times longer than search queries two years earlier.
Pichai rejects managed cannibalization as the wrong mindset: “The dilemma only exists if you treat it as a dilemma.” Mobile once raised similar monetization fears, while YouTube Shorts initially monetized far below long-form video. Google leaned into both experiences first: “Follow the user. All else will follow.”

2. AI search has reached an initial economic baseline

Friedberg cites recent court-testimony figures, including data from March, while noting that he does not know where the data came from: Gemini at 350 million monthly users, ChatGPT at 600 million, and Meta AI at 500 million. Pichai points to Gemini 2.5 Pro’s engagement lift, with Deep Research, Canvas, Audio Overviews, Veo 2 generation, and Gemini Live screen-sharing expanding the product.
Distribution is the rebuttal to app-only comparisons. Pichai argues that AI consumption already spans Search, YouTube, Cloud, Android-related Gemini experiences, and Gemini, with Search and AI Overviews potentially the most widely used generative-AI product. His test is narrower than market-share rhetoric: “If you innovate, are users responding and using it more?”
On serving economics, Pichai says the cost of a given AI query “has fallen dramatically in an 18-month time frame.” Google’s harder problem is latency because traditional Search is nearly instantaneous; infrastructure makes him confident cost itself will not determine whether the transition works.
Monetization has already reached a first checkpoint: ads shown with AI Overviews perform at the same baseline as results without them. Pichai expects improvement because “commercial information is also information” users seek when intent is present, though he preserves the timing hedge: “Some of it may take time.”

3. The infrastructure stack is Alphabet’s margin and capacity bet

Pichai says Google sits on the “Pareto frontier” of model performance and cost, with Gemini Flash serving as an industry workhorse. The advantage runs from subsea cables through data centers and seventh-generation TPUs to foundational research and products—a full stack built to train and serve Gemini at enormous scale.
Ironwood, the latest TPU generation discussed, has a single pod delivering more than 40 exaflops. Pichai also emphasizes that the chips are particularly strong for inference. That capability links directly to search economics: Google can introduce expensive model behavior across a mass-market product while continually lowering the infrastructure cost beneath it.
Alphabet plans $75 billion of capex in 2025, mostly for servers and data centers, with servers the largest portion. Half of compute spending goes toward Google Cloud; much of the rest supports DeepMind’s frontier work across language, images, video, and world models alongside Search, YouTube, and Gemini.
TPUs do not eliminate NVIDIA. Google trains Gemini internally on TPUs and serves it across its products that way, but also serves Gemini traffic on GPUs, deploys GPUs internally, and offers customers both. Pichai calls NVIDIA “a phenomenal company” with world-class R&D and software, while remaining “long-term committed to the TPU direction.”

4. Model progress is jagged, while the interface becomes ambient

Pichai borrows Andrej Karpathy’s phrase “artificial jagged intelligence”: progress pauses, then a paradigm breakthrough arrives. The frontier has moved through scaled pre-training, post-training, inference compute, and now agentic workflows; harder gains may distinguish elite teams, but researchers have not found a fundamental wall or a point where additional compute stops producing returns.
Google’s research extends beyond transformer-based LLMs into diffusion and other model families. The practical limits Pichai sees today are physical—data-center construction, power, and even finding enough electricians—not evidence that foundational model performance has stopped advancing.
Personal context is a potential product differentiator. With permission, Gmail, Calendar, Docs, YouTube, and Search could let Google provide more useful assistance; Pichai calls this a “differentiated innovation opportunity” but says the company still has to deliver it.
The interface endpoint is computing that demands less adaptation from humans. Natively multimodal models can take audio, vision, and language and remain in the user’s line of sight; Pichai thinks practical AR glasses are “a couple cycles away” from their 2006–2007 smartphone moment, with neural interfaces a longer-range possibility.

5. Competition enlarges the market—and DeepSeek resets China priors

Larry Page and Sergey Brin remain deeply engaged. Brin is “sitting and coding” alongside Gemini engineers, examining loss curves, model architecture, and post-training, while Pichai values three-way discussions because the founders are “very nonlinear thinkers” who anticipated moments like this 15 or 20 years ago.
Asked about Sam Altman, Elon Musk, Mark Zuckerberg, and Satya Nadella, Pichai avoids ranking them and jokes that only one invited him to a dance. His sharper observation is that Musk’s ability “to will future technologies into existence” is unparalleled, while the caliber of all four makes continued industry progress more likely.
Pichai rejects a winner-take-all frame: AI is “a much bigger opportunity landscape than all the previous technologies we have known combined.” As with the internet before Google existed, some eventual winners may be companies whose names are not yet known; execution, innovation, and talent matter more than today’s league table.
DeepSeek nevertheless changed expectations. Anyone following AI research and Chinese papers would already recognize the talent base, but Pichai says “all of us had to adjust our priors” about how close China was to the frontier. Hardware constraints drove DeepSeek’s efficiency innovations; internally, Google found Flash comparably efficient or, by some measures, better.

6. Power availability is already constraining AI deployment

Friedberg frames the geopolitical arithmetic: Elon Musk is discussing a terawatt of compute, roughly equivalent to US electricity-production capacity; he says that by 2040 the US may move from 1 to 2 terawatts while China moves from 3 to 8. That gap could shape where AI’s economic gains accrue.
Pichai agrees energy is the “most likely constraint for AI progress” and therefore GDP growth, but calls it an execution challenge rather than a physics barrier. The available portfolio includes solar-plus-batteries—which people “perpetually” underestimate—nuclear, geothermal, grid upgrades, transmission, and faster permitting.
Labor belongs in the bottleneck analysis: electricians leaving the workforce are colliding with rapidly rising data-center demand. Pichai says Google Cloud is already supply-constrained this year, with projects delayed by permitting and workforce shortages; continued AI investment will make those constraints more visible if deployments produce adequate economic returns.
Friedberg presses the 15-year downside: could China simply become the larger economy if US generation stays far behind? Pichai’s answer is an assumption, not a guarantee—capitalist solutions will respond through technologies such as small modular reactors and fusion, and the conversations will grow louder until capacity meets the moment.

7. Quantum and robotics carry explicit breakthrough windows

Quantum resembles AI around 2015, in Pichai’s pattern match. Because nature is fundamentally quantum, faithful large-scale simulations will ultimately require quantum computation; within roughly five years, he expects a useful calculation “far superior to classical computers,” and places Google’s frontier opportunity in a three-to-five-year window.
He acknowledges both technical risk and industry noise, comparing today’s quantum announcements with self-driving three years earlier, when many efforts appeared equivalent from outside but were not. Google’s route is to demonstrate useful algorithms and expose capability through Cloud, then let applications emerge that cannot be linearly predicted.
The analogy is Uber: smartphones, GPS, and payments made it possible, but no one could project Uber directly from those components. Quantum is similarly foundational; “we don’t know the algorithms yet.” Friedberg notes that limited access to quantum machines restricts experimentation, while Pichai says Google expects more exciting moments to share this year.
In robotics, Google “tried the application layer too early,” before AI materially improved physical systems. Google’s Gemini Robotics effort is developing vision-language-action models, and humanoid demonstrations have advanced enough that Pichai sometimes needs five seconds to decide whether footage is fake. He estimates a “magical moment” in two to three years, with Intrinsic effectively pursuing an Android-like layer for manufacturers.

8. Culture is being tightened around mission and shared technology

Pichai defends Google’s original perks as mechanisms for innovation: lunch put people together to exchange ideas; it was “not that we are trying to give lunch to people.” Employee agency still produces projects such as NotebookLM, but empowerment does not mean every internal voice represents the company—500 vocal employees can confuse the outside view.
His correction is mission focus: employees are not at Google “to resolve all our personal differences” but to innovate in service of the company’s purpose. DeepMind’s intensity reminds him of early Google, with teams working together in person five days a week or more; that passion is the “hardcoreness which matters.”
COVID was a major cultural discontinuity for a company designed around face-to-face exchange. Google restored a 3–2 hybrid model, created shared physical spaces for DeepMind teams in London and Mountain View, and recreated the Labs concept so roughly 10-person groups could pursue projects suited to that scale.
The AI talent market is fierce, but Pichai says Google is retaining critical talent, recruiting top PhDs, and seeing some former employees return. He is proud that Googlers have started more than 2,000 companies, creating a cycle of departures, returns, and acquisitions.
Alphabet, finally, is “not a holding company” that merely allocates capital to attractive assets. Quantum, Waymo, Cloud, Search, YouTube, Isomorphic, robotics, and other businesses are organized around underlying technology and R&D; some may eventually IPO, but common innovation is the organizing principle. Pichai’s example of an alternate path is Netflix, an acquisition Google once debated “super intensely”—perhaps realized elsewhere in the “multiverse.”

OpenAI, SBF & Perplexity: What VCs Know That You Don’t E191 | Small, Beautiful Opportunities Are Here: On the New Paradigm in This Wave of AI Agent Evolution